Post #1034: Trend to 2/26/2021, flat ex-Texas

Posted on February 27, 2021

Texas and some nearby states had severe difficulties in processing their COVID-19 case counts during the mid-February winter weather event.  That, combined with the use of seven day moving average data, means that their case counts are only now returning to normal.  You can see that reflected in the gray line on the first graph below, for the U.S. South Central states.  Including that data reporting effect makes it appear that new COVID cases are on the rise.

But excluding Texas (and Louisiana and Arkansas), cases are flat-to-down.  There’s still a sharp inflection point at 2/21/2021.   The decline in daily new case counts ceased at that point.  But so far, that’s been replaced with more-or-less stable new case counts.

And once the Texas (et al.) data reporting problems have worked their way through the data, that’s how the U.S. as a whole will appear.

The situation remains mixed, with about half of states showing declines in new case counts on any given day.  And the other half showing increases.

In keeping your thinking straight on this topic, you have to keep reminding yourself of the data lags and the use of a seven-day moving average.

If something perturbs data reporting, that begins to show up in the data immediately.  The use of a seven-day moving average then spreads that into a week-long anomaly.  In the past, “catch up” reporting by the states (a batch of old cases, all reported on a single day) generated what I termed speed bumps in the data.  (Search “speed bump” in Post #964, then Post #974).  With Texas, we had a lot of cases missing, so we got a combination speed dip/speed bump.  But the principle is the same.

If this little plateau is merely an after-effect of the recent cold snap on test reporting, then it’ll be over a couple of days from now.  Plausibly, a lot of states may have gotten backed up on case reporting during this period, but not to the obvious extent that Texas was.  If we’re just looking at the catch-up reporting, then that should be washing out of the numbers any day now.

By contrast, if something perturbs the actual underlying rate at which people are getting infected, that takes about 12 days (my estimate) to begin showing up in the data.  That’s due to the median five day lag between infection and onset of symptoms, and then the lags required to get to the doctor, get tested, have the test reports come back, and have that tabulated by the state.  (I wonder about the extent to which that has shrunk due to increased use of “instant” antigen-based tests, and the inclusion of antigen positives in the case reporting for all but a handful of U.S. states (e.g., Post #1016).  And then that is subject to the same smearing over time due to the use of a seven-day-moving-average. 

The upshot is that if you see an actual sea-change in the underlying infection rate, that’s due to something that occurred more than two weeks in the past.  So, if this little plateau reflects some sort of fundamental change in the infection rate, that change occurred on or around Super Bowl Sunday (2/5/2021).  But if this is a Super Bowl effect, it is notably absent from both the Tampa Bay and Kansas City infection rates (Post #1029).

Among other things, this lag between infection and reporting is why I find it so damned irritating that people still refer to the mythical post-Christmas surge in cases.  The seven-day moving average of reported new cases peaked on 1/8/2021.  Put aside that the peak merely aligned with pre-existing trend, so there was no “surge” above existing trend.  The key point is that you need to subtract ~16 days from that peak of reported infections, to find the peak of when those infections actually occurred.  If you do that, you realize that new infections actually began to fall well before Christmas day.  And that new infections fell sharply throughout the holiday season.  Which then showed up as the sharp decline in reported cases starting 1/8/2021.

Other that that, there’s nothing I can infer from the data.  Data reporting in a handful of states got hammered by the recent cold snap.   That left an artifact in the U.S. data, and we’re just now getting past that.

If that’s all it was, case counts will resume something like their prior downward trends.  If not, then not.

Based on my little bit of analysis of Tampa Bay and Kansas City, and the lack of a uniform effect across states, I’m betting that the timing of this, with respect to Super Bowl Sunday (plus 16 days), is just coincidence.

The next few days of data should be interesting.